utilitas 1998.2.60 → 1998.2.62

package/lib/alan.mjs

@@ -42,14 +42,14 @@ You may be provided with some tools(functions) to help you gather information an
 
 const _NEED = [
     '@anthropic-ai/sdk', '@anthropic-ai/vertex-sdk', '@google/generative-ai',
-    'js-tiktoken', 'ollama', 'OpenAI',
+    'js-tiktoken', 'OpenAI',
 ];
 
 const [
     OPENAI, GEMINI, CHATGPT, OPENAI_EMBEDDING, GEMINI_EMEDDING, OPENAI_TRAINING,
     OLLAMA, CLAUDE, GPT_4O_MINI, GPT_4O, GPT_O1, GPT_O3_MINI, GEMINI_20_FLASH,
     GEMINI_20_FLASH_THINKING, GEMINI_20_PRO, NOVA, EMBEDDING_001, DEEPSEEK_R1,
-    DEEPSEEK_R1_70B, MD_CODE, CHATGPT_REASONING, TEXT_EMBEDDING_3_SMALL,
+    DEEPSEEK_R1_70B, DEEPSEEK_R1_32B, MD_CODE, CHATGPT_REASONING, TEXT_EMBEDDING_3_SMALL,
     TEXT_EMBEDDING_3_LARGE, CLOUD_37_SONNET, AUDIO, WAV, CHATGPT_MINI,
     ATTACHMENTS, CHAT, OPENAI_VOICE, MEDIUM, LOW, HIGH, GPT_REASONING_EFFORT,
     THINK, THINK_STR, THINK_END, AZURE, TOOLS_STR, TOOLS_END, TOOLS, TEXT,
@@ -62,11 +62,12 @@ const [
         'OPENAI_TRAINING', 'OLLAMA', 'CLAUDE', 'gpt-4o-mini', 'gpt-4o', 'o1',
         'o3-mini', 'gemini-2.0-flash', 'gemini-2.0-flash-thinking-exp',
         'gemini-2.0-pro-exp', 'nova', 'embedding-001', 'deepseek-r1',
-        'deepseek-r1:70b', '```', 'CHATGPT_REASONING', 'text-embedding-3-small',
-        'text-embedding-3-large', 'claude-3-7-sonnet@20250219', 'audio', 'wav',
-        'CHATGPT_MINI', '[ATTACHMENTS]', 'CHAT', 'OPENAI_VOICE', 'medium',
-        'low', 'high', 'medium', 'think', '<think>', '</think>', 'AZURE',
-        '<tools>', '</tools>', 'tools', 'text', 'thinking', 'OK', 'function',
+        'deepseek-r1:70b', 'deepseek-r1:32b', '```', 'CHATGPT_REASONING',
+        'text-embedding-3-small', 'text-embedding-3-large',
+        'claude-3-7-sonnet@20250219', 'audio', 'wav', 'CHATGPT_MINI',
+        '[ATTACHMENTS]', 'CHAT', 'OPENAI_VOICE', 'medium', 'low', 'high',
+        'medium', 'think', '<think>', '</think>', 'AZURE', '<tools>',
+        '</tools>', 'tools', 'text', 'thinking', 'OK', 'function',
         'gpt-4.5-preview', 'redacted_thinking', 'gemma-3-27b-it',
         'AZURE OPENAI', 'ANTHROPIC', 'VERTEX ANTHROPIC', 'gemma3:27b',
         7680 * 4320, {}, 10, { log: true }, 'Alan', 'user', 'system',
@@ -369,6 +370,7 @@ const MODELS = {
 };
 
 MODELS[DEEPSEEK_R1_70B] = MODELS[DEEPSEEK_R1];
+MODELS[DEEPSEEK_R1_32B] = MODELS[DEEPSEEK_R1];
 MODELS[GEMMA327B] = MODELS[GEMMA_3_27B];
 
 for (const n in MODELS) {
@@ -565,11 +567,22 @@ const init = async (options = {}) => {
             };
             break;
         case OLLAMA:
+            // https://github.com/ollama/ollama/blob/main/docs/openai.md
+            const baseURL = 'http://localhost:11434/v1/';
             ais[id] = {
-                id, provider, model,
-                client: new (await need('ollama', { raw: true })).Ollama(options),
-                prompt: async (cnt, opts) => await promptOllama(id, cnt, opts),
+                id, provider, model, client: await OpenAI({
+                    baseURL, apiKey: 'ollama', ...options
+                }),
+                prompt: async (cnt, opts) => await promptOpenAI(id, cnt, opts),
             };
+            const phLog = m => log(`Ollama preheat: ${m?.message || m}`);
+            ignoreErrFunc(async () => {
+                phLog(await (await fetch(`${baseURL}completions`, {
+                    method: 'POST', body: JSON.stringify({
+                        model: model.name, prompt: '', keep_alive: -1
+                    })
+                })).text());
+            }, { log: phLog });
             break;
         default:
             throwError(`Invalid AI provider: ${options.provider || 'null'}.`);
@@ -638,14 +651,6 @@ const buildGptMessage = (content, options) => {
     return message;
 };
 
-const buildOllamaMessage = (content, options) => {
-    const message = String.isString(content) ? {
-        role: options?.role || user, content,
-    } : content;
-    assertContent(message.content);
-    return message;
-};
-
 const buildGeminiParts = (text, attachments) => {
     // Gemini API does not allow empty text, even you prompt with attachments.
     const message = [...text?.length || attachments?.length ? [{
@@ -695,8 +700,8 @@ const buildGeminiHistory = (text, options) => buildGeminiMessage(
     text, { ...options || {}, history: true }
 );
 
-const [getOpenAIClient, getGeminiClient, getOllamaClient, getClaudeClient]
-    = [OPENAI, GEMINI, OLLAMA, CLAUDE].map(
+const [getOpenAIClient, getGeminiClient, getClaudeClient]
+    = [OPENAI, GEMINI, CLAUDE].map(
         x => async options => await init({ ...provider(x), ...options })
     );
 
@@ -742,8 +747,7 @@ const packResp = async (resp, options) => {
     ])) && (audio = await convert(audio, {
         input: BUFFER, expected: BUFFER, ...options || {},
     }));
-    options?.jsonMode && !options?.delta && !options?.processing
-        && (json = parseJson(simpleText));
+    options?.jsonMode && !options?.delta && (json = parseJson(simpleText, null));
     if (options?.simple && options?.audioMode) { return audio; }
     else if (options?.simple && options?.jsonMode) { return json; }
     else if (options?.simple) { return simpleText; }
@@ -829,10 +833,6 @@ const buildPrompts = async (model, input, options = {}) => {
             systemPrompt = options.systemPrompt;
             prompt = buildClaudeMessage(content, { ...options, cache_control: true });
             break;
-        case OLLAMA:
-            systemPrompt = buildOllamaMessage(options.systemPrompt, _system);
-            prompt = buildOllamaMessage(content, options);
-            break;
         case GEMINI:
             const _role = { role: options.model === GEMMA_3_27B ? user : system };
             systemPrompt = buildGeminiHistory(options.systemPrompt, _role);
@@ -852,10 +852,6 @@ const buildPrompts = async (model, input, options = {}) => {
                     history.push(buildClaudeMessage(x.request, _user));
                     history.push(buildClaudeMessage(x.response, _assistant));
                     break;
-                case OLLAMA:
-                    history.push(buildOllamaMessage(x.request, _user));
-                    history.push(buildOllamaMessage(x.response, _assistant));
-                    break;
                 case GEMINI:
                     if (options.attachments?.length) { return; }
                     history.push(buildGeminiHistory(x.request, _user));
@@ -864,9 +860,14 @@ const buildPrompts = async (model, input, options = {}) => {
             }
         });
         switch (options.flavor) {
-            case CHATGPT: case CLAUDE: case OLLAMA:
+            case CHATGPT:
+                history = messages([
+                    systemPrompt, ...history, prompt,
+                    ...options.toolsResult?.length ? options.toolsResult : []
+                ]);
+                break;
+            case CLAUDE:
                 history = messages([
-                    ...options.flavor === CLAUDE ? [] : [systemPrompt],
                     ...history, prompt,
                     ...options.toolsResult?.length ? options.toolsResult : []
                 ]);
@@ -891,8 +892,7 @@ const buildPrompts = async (model, input, options = {}) => {
             content = trimTailing(trimTailing(content).slice(0, -1)) + '...';
         }
     }, model.maxInputTokens - options.attachments?.length * ATTACHMENT_TOKEN_COST);
-    if ([CHATGPT, OLLAMA].includes(options.flavor)
-        || options.model === GEMMA_3_27B) {
+    if ([CHATGPT].includes(options.flavor) || options.model === GEMMA_3_27B) {
         systemPrompt = null;
     }
     return { systemPrompt, history, prompt };
@@ -1010,6 +1010,7 @@ const promptOpenAI = async (aiId, content, options = {}) => {
     const resp = await client.chat.completions.create({
         model: azure ? undefined : options.model, ...history,
         ...options.jsonMode ? { response_format: { type: JSON_OBJECT } } : {},
+        ...provider === OLLAMA ? { keep_alive: -1 } : {},
         modalities, audio: options.audio || (
             modalities?.find?.(x => x === AUDIO)
             && { voice: DEFAULT_MODELS[OPENAI_VOICE], format: 'pcm16' }
@@ -1063,27 +1064,6 @@ const promptOpenAI = async (aiId, content, options = {}) => {
     return await packResp(event, options);
 };
 
-const promptOllama = async (aiId, content, options = {}) => {
-    const { client, model } = await getAi(aiId);
-    // https://github.com/ollama/ollama-js
-    // https://github.com/jmorganca/ollama/blob/main/examples/typescript-simplechat/client.ts
-    options.model = options?.model || model.name;
-    let [chunk, result] = [null, ''];
-    const { history }
-        = await buildPrompts(model, content, { ...options, flavor: OLLAMA });
-    const resp = await client.chat({
-        model: options.model, stream: true, ...history,
-    });
-    for await (chunk of resp) {
-        const delta = chunk.message.content || '';
-        result += delta;
-        delta && await streamResp({
-            text: options.delta ? delta : result,
-        }, options);
-    }
-    return await packResp({ text: result }, options);
-};
-
 const promptAnthropic = async (aiId, content, options = {}) => {
     const { client, model } = await getAi(aiId);
     let [
@@ -1458,7 +1438,7 @@ const talk = async (input, options) => {
         case CHATGPT: resp = await promptOpenAI(input, pmtOptions); break;
         case GEMINI: resp = await promptGemini(input, pmtOptions); break;
         case CLAUDE: resp = await promptAnthropic(input, pmtOptions); break;
-        case OLLAMA: resp = await promptOllama(input, pmtOptions); break;
+        // case OLLAMA: resp = await promptOllama(input, pmtOptions); break;
         case AZURE: resp = await promptAzure(input, pmtOptions); break;
         default: throwError(`Invalid AI engine: '${engine}'.`);
     }
@@ -1585,13 +1565,13 @@ const PREFERRED_ENGINES = [
     { client: GEMINI, func: promptGemini, multimodal: 1 },
     { client: CLAUDE, func: promptAnthropic, multimodal: 2 },
     // { client: AZURE, func: promptAzure, multimodal: 3 },
-    { client: OLLAMA, func: promptOllama, multimodal: 99 },
+    // { client: OLLAMA, func: promptOllama, multimodal: 99 },
 ]; // keep gpt first to avoid gemini grounding by default
 
 export default init;
 export {
     ATTACHMENT_TOKEN_COST, CLOUD_37_SONNET, CODE_INTERPRETER, DEEPSEEK_R1,
-    DEEPSEEK_R1_70B, DEFAULT_MODELS,
+    DEEPSEEK_R1_32B, DEEPSEEK_R1_70B, DEFAULT_MODELS,
     EMBEDDING_001,
     FUNCTION, GEMINI_20_FLASH, GEMINI_20_FLASH_THINKING, GPT_45, GPT_4O, GPT_4O_MINI, GPT_O1, GPT_O3_MINI, INSTRUCTIONS, MODELS,
     OPENAI_VOICE, RETRIEVAL,
@@ -1619,7 +1599,6 @@ export {
     prompt, promptOpenAI,
     promptAnthropic,
     promptGemini,
-    promptOllama,
     resetSession,
     tailGptFineTuningEvents,
     talk,

package/lib/manifest.mjs

@@ -1,7 +1,7 @@
 const manifest = {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.60",
+    "version": "1998.2.62",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",
@@ -53,7 +53,6 @@ const manifest = {
         "node-mailjet": "^6.0.8",
         "node-polyfill-webpack-plugin": "^4.1.0",
         "office-text-extractor": "^3.0.3",
-        "ollama": "^0.5.14",
         "openai": "^4.87.3",
         "pdfjs-dist": "^4.10.38",
         "pg": "^8.14.0",

package/lib/utilitas.mjs

@@ -650,8 +650,10 @@ const checkInterval = (itv, sed) =>
 
 const ignoreErrFunc = async (func, options) => {
     const run = async () => {
-        try { return await func(...options?.args || []) }
-        catch (err) { options?.log && console.error(err); }
+        try { return await func(...options?.args || []) } catch (err) {
+            if (Function.isFunction(options?.log)) { options.log(err); }
+            else if (options?.log) { console.error(err); }
+        }
     };
     if (options?.await) { await timeout(options.await); return await run(); }
     else if (options?.wait) { return setTimeout(run, options.wait); }

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.60",
+    "version": "1998.2.62",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",
@@ -64,7 +64,6 @@
         "node-mailjet": "^6.0.8",
         "node-polyfill-webpack-plugin": "^4.1.0",
         "office-text-extractor": "^3.0.3",
-        "ollama": "^0.5.14",
         "openai": "^4.87.3",
         "pdfjs-dist": "^4.10.38",
         "pg": "^8.14.0",